Fabien Roger comments on An issue with training schemers with supervised fine-tuning